The impact of relative radiometric calibration on the accuracy of kNN-predictions of forest attributes

The k-nearest-neighbour (kNN) algorithm is widely applied for the estimation of forest attributes using remote sensing data. It requires a large amount of reference data to achieve satisfactory results. Usually, the number of available reference plots for the kNN-prediction is limited by the size of the area covered by a terrestrial reference inventory and remotely sensed imagery collected from one overflight. The applicability of kNN could be enhanced if adjacent images of different acquisition dates could be used in the same estimation procedure. Relative radiometric calibration is a prerequisite for this. This study focuses on two empirical calibration methods. They are tested on adjacent LANDSAT TM scenes in Austria. The first, quite conventional one is based on radiometric control points in the overlap area of two images and on the determination of transformation parameters by linear regression. The other, recently developed method exploits the kNN-cross-validation procedure. Performance and applicability of both methods as well as the impact of phenology are discussed.     

Object-based analysis and change detection of major wetland cover types and their classification uncertainty during the low water period at Poyang Lake, China

Productive wetland systems at land-water interfaces that provide unique ecosystem services are challenging to study because of water dynamics, complex surface cover and constrained field access. We applied object-based image analysis and supervised classification to four 32-m Beijing-1 microsatellite images to examine broad-scale surface cover composition and its change during November 2007-March 2008 low water season at Poyang Lake, the largest freshwater lake-wetland system in China (> 4000 km²). We proposed a novel method for semi-automated selection of training objects in this heterogeneous landscape using extreme values of spectral indices (SIs) estimated from satellite data. Dynamics of the major wetland cover types (Water, Mudflat, Vegetation and Sand) were investigated both as transitions among primary classes based on maximum membership value, and as changes in memberships to all classes even under no change in a primary class. Fuzzy classification accuracy was evaluated as match frequencies between classification outcome and a) the best reference candidate class (MAX function) and b) any acceptable reference class (RIGHT function). MAX-based accuracy was relatively high for Vegetation (≥ 90%), Water (≥ 82%), Mudflat (≥ 76%) and the smallest-area Sand (≥ 75%) in all scenes; these scores improved with the RIGHT function to 87-100%. Classification uncertainty assessed as the proportion of fuzzy object area within a class at a given fuzzy threshold value was the highest for all classes in November 2007, and consistently higher for Mudflat than for other classes in all scenes. Vegetation was the dominant class in all scenes, occupying 41.2-49.3% of the study area. Object memberships to Vegetation mostly declined from November 2007 to February 2008 and increased substantially only in February-March 2008, possibly reflecting growing season conditions and grazing. Spatial extent of Water both declined and increased during the study period, reflecting precipitation and hydrological events. The “fuzziest” Mudflat class was involved in major detected transitions among classes and declined in classification accuracy by March 2008, representing a key target for finer-scale research. Future work should introduce Vegetation sub-classes reflecting differences in phenology and alternative methods to discriminate Mudflat from other classes. Results can be used to guide field sampling and top-down landscape analyses in this wetland.     

Analysis of monotonic greening and browning trends from global NDVI time-series

Remotely sensed vegetation indices are widely used to detect greening and browning trends; especially the global coverage of time-series normalized difference vegetation index (NDVI) data which are available from 1981. Seasonality and serial auto-correlation in the data have previously been dealt with by integrating the data to annual values; as an alternative to reducing the temporal resolution, we apply harmonic analyses and non-parametric trend tests to the GIMMS NDVI dataset (1981-2006). Using the complete dataset, greening and browning trends were analyzed using a linear model corrected for seasonality by subtracting the seasonal component, and a seasonal non-parametric model. In a third approach, phenological shift and variation in length of growing season were accounted for by analyzing the time-series using vegetation development stages rather than calendar days. Results differed substantially between the models, even though the input data were the same. Prominent regional greening trends identified by several other studies were confirmed but the models were inconsistent in areas with weak trends. The linear model using data corrected for seasonality showed similar trend slopes to those described in previous work using linear models on yearly mean values. The non-parametric models demonstrated the significant influence of variations in phenology; accounting for these variations should yield more robust trend analyses and better understanding of vegetation trends.     

The effect of the temporal resolution of NDVI data on season onset dates and trends across Canadian broadleaf forests

Satellite remote sensing has the potential to contribute to plant phenology monitoring at spatial and temporal scales relevant for regional and global scale studies. Historically, temporal composites of satellite data, ranging from 8 days to 16 days, have been used as a starting point for satellite-derived phenology data sets. In this study we assess how the temporal resolution of such composites affects the estimation of the start of season (SOS) by: 1) calibrating a relationship between satellite derived SOS with in situ leaf unfolding (LU) of trembling aspen (Populus tremuloides) across Canada and 2) quantifying the sensitivity of calibrated satellite SOS estimates and trends, over Canadian broadleaf forests, to the temporal resolution of NDVI data. SOS estimates and trends derived from daily NDVI data were compared to SOS estimates and trends derived from multiday NDVI composites that retain the exact date of the maximum NDVI value or that assume the midpoint of the multiday interval as the observation date. In situ observations of LU dates were acquired from the PlantWatch Canada network. A new Canadian database of cloud and snow screened daily 1-km resolution National Oceanic and Atmospheric Administration advanced very high resolution radiometer surface reflectance images was used as input satellite data. The mean absolute errors of SOS dates with respect to in situ LU dates ranged between 13 and 40 days. SOS estimates from NDVI composites that retain the exact date of the maximum NDVI value had smaller errors (~ 13 to 20 days). The sensitivity analysis reinforced these findings: SOS estimates from NDVI composites that use the exact date had smaller absolute deviations from the LU date (0 to − 5 days) than the SOS estimates from NDVI composites that use the midpoint (− 2 to − 27 days). The SOS trends between 1985 and 2007 were not sensitive to the temporal resolution or compositing methods. However, SOS trends at individual ecozones showed significant differences with the SOS trends from daily NDVI data (Taiga plains and the Pacific maritime zones). Overall, our results suggest that satellite based estimates of vegetation green-up dates should preferably use sub-sampled NDVI composites that include the exact observation date of the maximum NDVI to minimize errors in both, SOS estimates and SOS trend analyses. For trend analyses alone, any of the compositing methods could be used, preferably with composite intervals of less than 28 days. This is an important finding, as it suggests that existing long-term 10-day or 15-day NDVI composites could be used for SOS trend analyses over broadleaf forests in Canada or similar areas. Future studies will take advantage of the growing in situ phenology networks to improve the validation of satellite derived green-up dates.     

Characterization of seasonal variation of forest canopy in a temperate deciduous broadleaf forest, using daily MODIS data

In this paper, we present an improved procedure for collecting no or little atmosphere- and snow-contaminated observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor. The resultant time series of daily MODIS data of a temperate deciduous broadleaf forest (the Bartlett Experimental Forest) in 2004 show strong seasonal dynamics of surface reflectance of green, near infrared and shortwave infrared bands, and clearly delineate leaf phenology and length of plant growing season. We also estimate the fractions of photosynthetically active radiation (PAR) absorbed by vegetation canopy (FAPAR_canopy), leaf (FAPAR_leaf), and chlorophyll (FAPAR_chl), respectively, using a coupled leaf-canopy radiative transfer model (PROSAIL-2) and daily MODIS data. The Markov Chain Monte Carlo (MCMC) method (the Metropolis algorithm) is used for model inversion, which provides probability distributions of the retrieved variables. A two-step procedure is used to estimate the fractions of absorbed PAR: (1) to retrieve biophysical and biochemical variables from MODIS images using the PROSAIL-2 model; and (2) to calculate the fractions with the estimated model variables from the first step. Inversion and forward simulations of the PROSAIL-2 model are carried out for the temperate deciduous broadleaf forest during day of year (DOY) 184 to 201 in 2005. The reproduced reflectance values from the PROSAIL-2 model agree well with the observed MODIS reflectance for the five spectral bands (green, red, NIR₁, NIR₂, and SWIR₁). The estimated leaf area index, leaf dry matter, leaf chlorophyll content and FAPAR_canopy values are close to field measurements at the site. The results also showed significant differences between FAPAR_canopy and FAPAR_chl at the site. Our results show that MODIS imagery provides important information on biophysical and biochemical variables at both leaf and canopy levels.     

Radar remote sensing of the spring thaw transition across a boreal landscape

The seasonal transition of the boreal forest between frozen and non-frozen conditions affects a number of ecosystem processes that cycle between winter dormant and summer active states. The relatively short K_u-band wavelength (2.14 cm) of the space-borne NASA scatterometer (NSCAT) is sensitive to changes in dielectric properties, associated with large-scale changes in the relative abundance and phase (frozen or thawed) of canopy and surface water. We used a temporal change detection analysis of NSCAT daily radar backscatter measurements to characterize the 1997 seasonal spring thaw transition period across the 10⁶ km² BOREAS study region of central Canada. In the spring, air temperature transitions from frozen to non-frozen conditions and surface observations of seasonal snow cover depletion were generally coincident with decreases in radar backscatter of more than 2.9 dB, regardless of regional landcover characteristics. We used a temporal classification of NSCAT daily differences from 5-day smoothed backscatter values to derive three simple indices describing the initiation, primary event and completion of the spring thaw transition period. Several factors had a negative impact on the relative accuracy of NSCAT-based results, including periodic gaps in NSCAT daily time-series information and a large (i.e., >2 cm day⁻¹) spring rainfall event. However, these results were generally successful in capturing the seasonal transition of the region from frozen to non-frozen conditions, based on comparisons with regional weather station network information. These results illustrate the potential for improved assessment of springtime phenology and associated ecosystem dynamics across high latitude regions, where field based and optical remote-sensing methods are substantially degraded by frequent cloud cover, low solar illumination and sparse surface weather station networks.     

Satellite radar remote sensing of seasonal growing seasons for boreal and subalpine evergreen forests

We evaluated whether satellite radar remote sensing of landscape seasonal freeze-thaw cycles provides an effective measure of active growing season timing and duration for boreal and subalpine evergreen forests. Landscape daily radar backscatter measurements from the SeaWinds scatterometer on-board the QuikSCAT satellite were evaluated across a regional network of North American coniferous forest sites for 2000 and 2001. Radar remote sensing measurements of the initiation and length of the growing season corresponded closely with both site measurements and ecosystem process model (BIOME-BGC) simulations of these parameters because of the sensitivity of the K_u-band scatterometer to snow cover freeze-thaw dynamics and associated linkages between growing season initiation and the timing of seasonal snowmelt. In contrast, remote sensing estimates of the timing of growing season termination were either weakly or not significantly associated with site measurements and model simulation results, due to the relative importance of light availability and other environmental controls on stand phenology in the fall. Regional patterns of estimated annual net primary production (NPP) and component photosynthetic and autotrophic respiration rates for the evergreen forest sites also corresponded favorably with remote sensing estimates of the seasonal timing of spring thaw and associated growing season length, indicating the importance of these parameters in determining spatial and temporal patterns of NPP and the potential utility of satellite radar remote sensing for regional monitoring of the terrestrial biosphere.     

Integration of MODIS-derived metrics to assess interannual variability in snowpack, lake ice, and NDVI in southwest Alaska

Impacts of global climate change are expected to result in greater variation in the seasonality of snowpack, lake ice, and vegetation dynamics in southwest Alaska. All have wide-reaching physical and biological ecosystem effects in the region. We used Moderate Resolution Imaging Spectroradiometer (MODIS) calibrated radiance, snow cover extent, and vegetation index products for interpreting interannual variation in the duration and extent of snowpack, lake ice, and vegetation dynamics for southwest Alaska. The approach integrates multiple seasonal metrics across large ecological regions.Throughout the observation period (2001-2007), snow cover duration was stable within ecoregions, with variable start and end dates. The start of the lake ice season lagged the snow season by 2 to 3 months. Within a given lake, freeze-up dates varied in timing and duration, while break-up dates were more consistent. Vegetation phenology varied less than snow and ice metrics, with start-of-season dates comparatively consistent across years. The start of growing season and snow melt were related to one another as they are both temperature dependent. Higher than average temperatures during the El Niño winter of 2002-2003 were expressed in anomalous ice and snow season patterns. We are developing a consistent, MODIS-based dataset that will be used to monitor temporal trends of each of these seasonal metrics and to map areas of change for the study area.     

Interannual vegetation phenology estimates from global AVHRR measurements: Comparison with in situ data and applications

In a previous paper, we described a procedure to correct the directional effects in AVHRR reflectance time series. The corrected measurements show much less high frequency variability than their original counterparts, which makes them suitable to study vegetation dynamics. The time series are used here to estimate the start and ending dates of the growing season for 18 years from 1982 to 1999. We focus on the interannual variations of these phenological parameters.A database of in situ phenology observations is used to quantify the accuracy of the satellite-based estimates. Although based on a limited sampling of the Northern mid and high latitudes, the comparison indicates that i) the satellite phenological product contains meaningful information on interannual onset anomalies; ii) there is a higher degree of consistency over regions covered by Broadleaf Forests, Grasses and cereal Crops than over those covered by Needleleaf Forests or Savannas; and iii) the satellite phenological product is of lower quality in regions with mountainous terrain. In favorable conditions, interannual variations of the onset are captured with an accuracy of a few days.As this satellite-derived product captures the interannual onset variability at ground-truth sites, we confidently use it to larger scales studies. Mapped at a continental scale, the onset anomalies show coherent patterns at the regional (≈ 1000 km) scale for the mid and high latitudes of the Northern hemisphere, which is consistent with a meteorological forcing. In the tropics, there is larger spatial heterogeneity, which suggests more complex controls of the phenology. The relation between vegetation phenology and climate is further investigated over Europe by comparing the variability of the satellite-derived vegetation onset and that of the winter North Atlantic Oscillation index, at a fine spatial scale. The strong correlations observed confirm that this climate forcing parameter explains most of the onset variability over a large fraction of Northern Europe (earlier onsets for positive winter NAO), with lower impact towards the south and opposite effects around the Mediterranean basin. The NAO has a predictive character as the January-February NAO index is strongly correlated with the vegetation onset that occurs around April in Northern Europe.     

A curve fitting procedure to derive inter-annual phenologies from time series of noisy satellite NDVI data

Annual, inter-annual and long-term trends in time series derived from remote sensing can be used to distinguish between natural land cover variability and land cover change. However, the utility of using NDVI-derived phenology to detect change is often limited by poor quality data resulting from atmospheric and other effects. Here, we present a curve fitting methodology useful for time series of remotely sensed data that is minimally affected by atmospheric and sensor effects and requires neither spatial nor temporal averaging. A two-step technique is employed: first, a harmonic approach models the average annual phenology; second, a spline-based approach models inter-annual phenology. The principal attributes of the time series (e.g., amplitude, timing of onset of greenness, intrinsic smoothness or roughness) are captured while the effects of data drop-outs and gaps are minimized. A recursive, least squares approach captures the upper envelope of NDVI values by upweighting data values above an average annual curve. We test this methodology on several land cover types in the western U.S., and find that onset of greenness in an average year varied by less than 8 days within land cover types, indicating that the curve fit is consistent within similar systems. Between 1990 and 2002, temporal variability in onset of greenness was between 17 and 35 days depending on the land cover type, indicating that the inter-annual curve fit captures substantial inter-annual variability. Employing this curve fitting procedure enhances our ability to measure inter-annual phenology and could lead to better understanding of local and regional land cover trends.